The full Phase 4 stack. Each metric was face-validity tested before it shipped — if AVI said a bench player out-impacted Wilson, we threw it out. The capstone is W_AVI, a composite of 6 SHIP_CLEAN metrics that stabilizes the underlying noise: any single metric has year-over-year correlation under +0.2; the composite is +0.77. Multi-metric substrate cancels measurement noise and compounds true signal.
Top 500 by projected minutes. Click any column header to sort; click a metric name to open its ranking page. Empty cells (—) mean the player didn't qualify under that metric's minimum-sample threshold.
Every Resolve WNBA advanced metric, with what it measures and how to read it. Each metric had to pass a face-validity gate before shipping — if a metric ranked someone obviously wrong, it got cut. Grouped by side (offense / defense / two-way / team).
How much opponents miss at the rim with this player on the floor.
What it measures: Opponent FG% at distance ≤5ft, on-vs-off court delta. Restricted to players with adequate rim attempts faced (≥100 on-court).
How to read it: Negative = elite (opponents shoot worse). Magbegor 2024 -0.147 is elite-tier; her 2025 +0.05 reversal is the league's biggest cross-batch decline (confirmed by 6 independent metrics in W_AVI).
The MVP-of-the-stack capstone — composite of 6 SHIP_CLEAN metrics.
What it measures: Composite value score combining O-RAPM, D-RAPM, NetRtg on-off, W_RPV, USG, and PPG into one number per player-season. Each input z-scored and weighted by √reliability, then scaled 0-100.
How to read it: 100 = league best, 50 = league average. A'ja Wilson 100.0 both 2024 and 2025 (unanimous MVP). YoY ρ=+0.769 — composite stability is the methodological punchline: any single metric is noisy, the stack is not.
Ridge-regularized lineup impact per-100 possessions.
What it measures: Ridge α=300 regression of per-possession net rating on lineup membership. The standard 'impact' metric — McBride +7.15, Stewart +5.60, Wilson +4.24 lead the league.
How to read it: Per-100 net rating impact above replacement. Stable enough at α=300 for half-season samples; ρ=+0.648 with raw NetRtg on-off.
Two-way star detector.
What it measures: Splits RAPM into separate offensive and defensive Ridge fits. Surfaces players who carry on both ends vs role specialists.
How to read it: Two-way stars: Stewart (O+3.95 / D+5.83), Bueckers (O+4.45 / D+4.62 rookie). High on one only = specialist. Orthogonality ρ(O,D)=+0.087 confirms they're truly independent axes.
Raw on/off + the four shooting/turnover/free-throw decomposition.
What it measures: Net rating with player on vs off the floor, plus eFG/TOV%/FTr decomposition for both sides. The unregularized baseline RAPM is built against.
How to read it: Wilson 2025 +24.41 NetRtg on-off leads. ρ_eFG_Net=+0.577 confirms shooting dominates four-factor contribution to net rating.
Which teammates' contributions rise together vs trade off.
What it measures: For each pair of teammates, the correlation of their per-game event shares when both are on the floor vs solo. Positive = activate together, negative = shift-share.
How to read it: Caitlin Clark's gravity surfaces here — Bonner-with-Clark coupling -0.099 means Bonner's share collapses when Clark plays. Negative coupling isn't bad — it's structural division of labor.
Last-5-min, score within 5 — net rating when it's on the line.
What it measures: On-court net rating restricted to clutch time (final 5 min, score margin ≤5). Volatile by design — clutch samples are small.
How to read it: ρ_clutch_vs_full=+0.051 — clutch is its OWN dimension, not a regular-time predictor. Top: Erica Wheeler +53.4 (huge sample caveat), Wallace +26.4. Use as a tiebreaker, not a primary metric.